Flexible parametric models for survival-time data with background mortality
strsrcs [varlist] [if exp] [in range] [, model_complexity bhazard(varname) scale(hazard|odds) strata(strata_varlist) noconstant snoconstant orthog eform inits(name) maximize_options ]
where model_complexity is one of
{ df(#)} knots(knotlist)} }
strsrcs is for use with survival-time data; see help st. You must stset your data before using strsrcs; see help stset.
strsrcs shares the features of all estimation commands; see help estcom.
The syntax of predict following strsrcs is
predict newvarname [if exp] [in range] [, statistic ci level(value)
where statistic is
survival - predicted relative survival
hazard - predicted excess hazard rate
hratio - predicted excess hazard rate ratio (NOTE:Only valid when one covariate is modelled)
These statistics are available both in and out of sample; type "predict ... if e(sample) ..." if wanted only for the estimation sample.
Predictions are obtained using the delta method implemented using predictnl.
Description
strsrcs fits spline-based distributional models to right censored data taking background mortality into account. varlist is a set of covariates.
Options for strsrcs
df(#) specifies the degrees of freedom for the natural spline function. # must be between 2 and 6. The knots() option is not applicable and the knots are placed at the following centiles of the distribution of the uncensored event [i.e. where _d==1] log times:
--------------------------- df Centile positions --------------------------- 1 (no knots) 2 50 3 33 67 4 25 50 75 5 20 40 60 80 6 17 33 50 67 83 >6 (not allowed) ---------------------------
knots(knotlist) knot placement that defines the internal knot positions for the spline. The values in knotlist are taken to be centile positions in the distribution of the uncensored event [i.e. where _d==1] log times.
bhazard(varname) gives the variable name for the baseline hazard at death/censoring. This option is compulsary, but a variable containing only zeros can be specified if 'standard' survival models are required as in stpm.
scale(hazard|odds) is not optional and specifies the scale of the model. The hazard and odds options fit models on the scale of the log cumulative hazard or the log cumulative odds of failure, respectively.
strata(strata_varlist) stratifies the spline functions according to the variables in strat_varlist. This allows time-dependant effects for a variety of covariates and and is of particular use in assessing the assumption of proportional excess hazards or proportional odds.
noconstant specifies that a constant term is not included in the baseline part of the model.
snoconstant specifies that a constant term is not included for the spline terms.
orthog creates orthogonalized basis functions. All basis functions higher than the first (linear) function are uncorrelated and have mean 0 and standard deviation 1. The linear function is also uncorrelated with the higher-basis functions.
eform displays the exponentiated coefficients and corresponding standard errors and confidence intervals.
inits(name) specifies a user defined set of initial values, where name is the name of the matrix where these values are stored.
level(#) specifies the confidence level, in percent, for confidence intervals. The default is level(95) or as set by set level; see help level.
maximize_options control the maximization process; see help maximize.
Examples
. strsrcs sex, bhazard(rate) df(2) scale(odds) [assess sex on the odds scale assuming proportional odds] . strsrcs, bhazard(rate) df(2) scale(odds) strata(sex) [assess sex on the odds scale assuming sex is dependent on time > ] . strsrcs age, bhaz(rate) df(6) sc(hazard) strata(sex) [assess age and sex on the hazard scale assuming age has propor > tional hazards and sex is time dependent] . predict s, survival ci [output the relative survival estimate (with confidence interva > l) into new variable s(s_lci & s_uci)] . predict haz, h ci level(99) [output the excess hazard (with confidence interval at the 99% > level) into new variable haz(haz_lci & haz_uci)]
Auxiliary ado-files
rcs strsrcs_mlo strsrcs_mlh strsrcs_pred
Author
Chris Nelson, University of Leiceser, UK. cn46@le.ac.uk
References
Nelson, C. P., Lambert, P. C., Squire, I. S. and Jones, D. R. 2007. Flexibl > e parametric models for relative survival, with application in coronary heart disease. Statistics in Medicine 26: 5486-5498.
Royston, P. 2001. Flexible alternatives to the Cox model, and more. Stata > Journal 1: 1-28.
Royston, P. and M. K. B. Parmar. 2002. Flexible proportional-hazards and pr > oportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects. S > tatistics in Medicine 21: 2175-2197.
Also see